Single-Objective Optimization for Architecture Guilherme Ilunga Thesis to obtain the Master of Science Degree in Information Systems and Computer Engineering Supervisors: Prof. Dr. Antonio´ Paulo Teles de Menezes Correia Leitao˜ February 2019 Agradecimentos Agradec¸o ao meu orientador. O professor Antonio´ Leitao,˜ ao longo de todo este tempo, mais do que pensar apenas no trabalho, pensou em mim. Dedicou tempo mesmo quando nao˜ tinha, para eu poder cumprir os meus objetivos. Agradec¸o ao Grupo de Arquitetura e Computadores, grupo com o qual trabalhei durante o desen- volvimento desta tese. Muito obrigado Bruno Ferreira, Catarina Belem,´ Francisco Loio, Inesˆ Caetano, Inesˆ Pereira, Renata Castelo Branco, Rita Aguiar, Sofia Feist, Sofia Sousa, e aos restantes membros do grupo. Agradec¸o aos professores Lu´ıs Rodrigues, Joao˜ Barreto, e Miguel Pardal. Durante a realizac¸ao˜ desta tese, tive a oportunidade de trabalhar como bolseiro de apoio a` docenciaˆ para as cadeiras de Sistemas Operativos e Sistemas Distribu´ıdos. Muito obrigado a estes professores, e aos restantes docentes destas cadeiras, por me terem permitido trabalhar enquanto realizava a tese. Agradec¸o aos meus amigos. Os meus amigos do Instituto Superior Tecnico,´ Carolina Fernandes, Catarina Belem,´ Claudia´ Belem,´ Filipe Magalhaes,˜ Gonc¸alo Rodrigues, Nuno Afonso, Telma Correia, Valentyn Hulevych, e os meus amigos do Externato Frei Lu´ıs de Sousa, Ana Rita Bello, Ana Teresa Martins, Francisco Henriques, Inesˆ Franco, Paulo Santos, e Tomas´ Fernandes. Mesmo tendo as suas proprias´ teses ou outros deveres, todas estas pessoas me apoiaram no meu percurso. Por fim, agradec¸o a` minha fam´ılia, cujo apoio incondicional foi a base de todo este trabalho. Muito obrigado aos meus pais, avos,´ e tios. Abstract The focus on efficiency has grown over recent years and, nowadays, it is critical that buildings exhibit good performance regarding different criteria. This need prompts architects to explore (1) algorithmic design approaches, which allow the generation of several design variations, (2) analysis tools, to eval- uate a design’s performance, and (3) optimization algorithms, to find the best performing variation of a design. Many optimization algorithms exist, and not all of them are adequate for a specific problem, however Genetic Algorithms are frequently the first and only option in architectural optimization. This may be because existing optimization frameworks require the usage of specific design or analysis tools and only offer a small subset of optimization algorithms, which leads to the simplest algorithm being the usual choice. This dissertation studies existing approaches and optimization algorithms for architecture. It also proposes a framework for architectural optimization that includes several types of algorithms and a prototype implementation. This prototype is tested using case studies, and the results show that Ge- netic Algorithms perform poorly, while other algorithms achieve better results. However, they also show that no algorithm is consistently better than the others, which suggests that a framework with several different types of algorithms should be used. Keywords Black-box Optimization; Derivative-free Optimization; Architectural Optimization; Performance-based Design; Algorithmic Design. iii Resumo O foco em eficienciaˆ tem crescido nos ultimos´ anos e, atualmente, e´ cr´ıtico que os edif´ıcios demon- strem um bom desempenho em diferentes criterios.´ Esta necessidade leva os arquitetos a explorar (1) design algor´ıtmico, que permite a gerac¸ao˜ de multiplas´ variac¸oes˜ do design, (2) ferramentas de analise,´ para avaliar o desempenho de um design, e (3) algoritmos de optimizac¸ao,˜ para encontrar a variac¸ao˜ do design com o melhor desempenho. Existem varios´ algoritmos de optimizac¸ao˜ e nem todos sao˜ adequados para um determinado problema, contudo Algoritmos Geneticos´ sao˜ frequentemente a primeira e unica´ opc¸ao˜ em optimizac¸ao˜ arquitetural. Isto ocorre porque apesar de existirem algumas frameworks de optimizac¸ao,˜ a maioria delas necessita de ferramentas de design ou analise´ e apenas oferecem um pequeno subconjunto de algoritmos, o que leva a que o algoritmo mais simples seja a escolha usual. Esta dissertac¸ao˜ estuda as abordagens existentes e os algoritmos de optimizac¸ao˜ para arquitetura. Tambem´ propoe˜ uma framework para otimizac¸ao˜ arquitetural que inclui varios´ tipos de algoritmos e que implementa num prototipo.´ O prototipo´ foi testado utilizando casos de estudo, e os resultados demonstram que os Algoritmos Geneticos´ nao˜ temˆ um bom desempenho, enquanto que out- ros algoritmos obtemˆ melhores resultados. Contudo, os resultados tambem´ demonstram que nenhum algoritmo e´ consistentemente melhor que os restantes, o que sugere que uma framework com varios´ algoritmos de diferentes tipos deve ser utilizada. Palavras Chave Optimizac¸ao˜ de Caixa Preta; Optimizac¸ao˜ Sem Derivadas; Optimizac¸ao˜ Arquitetural; Design Baseado em Performance; Design Algoritmico v Contributions During the development of this dissertation, two scientific articles were published: • Case Studies on the Integration of Algorithmic Design Processes in Traditional Design Workflows, published in the 23rd International Conference of the Association for Computer-Aided Architectural Design Research in Asia [Caetano et al., 2018] • Derivative-free Methods for Structural Optimization, published in the 36th Education and Research in Computer-Aided Architectural Design in Europe Conference [Ilunga and Leitao,˜ 2018] vii viii Contents 1 Introduction 1 1.1 Design Stage...........................................5 1.2 Analysis Stage..........................................8 1.3 Optimization Stage........................................ 13 1.4 Dissertation Objectives..................................... 15 2 Optimization 17 2.1 Optimization Algorithms..................................... 19 2.1.1 Sampling algorithms................................... 21 2.1.2 Direct-search algorithms................................. 21 2.1.3 Metaheuristics...................................... 23 2.1.4 Model-based algorithms................................. 25 2.2 Optimization in Architecture................................... 27 3 Optimization Framework for Architecture 29 3.1 Framework Description..................................... 31 3.2 Prototype Description...................................... 32 4 Evaluation 39 4.1 Test Functions.......................................... 41 4.1.1 Himmelblau Function.................................. 42 4.1.2 Levy Number 13 Function................................ 45 4.1.3 Rastringin Function................................... 47 4.2 Case Studies........................................... 50 4.2.1 An Urban Museum.................................... 50 4.2.2 Space Frame Optimization............................... 51 5 Discussion 55 5.1 Conclusions............................................ 57 5.2 Future Work............................................ 59 ix x List of Figures 1.1 A representation of the different panels of the Soumaya Museum in Mexico City. The number of different panels was optimized to be minimal as a cost reduction technique...3 1.2 The 30 St. Mary Axe building in London, also known as the Gherkin, designed by Foster and Partners, together with the Arup Group...........................4 1.3 A space frame with a sinusoidal shape.............................6 1.4 Original design of the Astana National Library (left) and variations created withAD (center and right). Adapted from [Branco and Leitao,˜ 2017]......................7 1.5 The complete Grasshopper algorithm for the Hangzhou Tennis Center. Adapted from [Miller, 2011]............................................8 1.6 An example of using Rosetta to generate a design on several design tools. Adapted from [Branco and Leitao,˜ 2017]..................................9 1.7 An example of the radiation map obtained by using the Radiance analysis tool on two variations of the Astana National Library. Adapted from [Aguiar et al., 2017]......... 10 1.8 An example of the tension map obtained by using the Robot analysis tool on a space frame. 10 1.9 Diferent models of the Shenzhen Bao’an International Airport Terminal 3. Geometrical 3D model (left), analytical model for radiation analysis (center) and analytical model for structural analysis (right). Adapted from [Aguiar et al., 2017]................. 11 1.10 The ADA workflow, where theAD tool is capable of generating geometrical or analytical models from the same algorithm................................. 12 1.11 The Meiso no Mori Crematorium in Japan, designed by Toyo Ito and Associates. The building’s roof has a non-uniform shape, similar to a cloud, and its supporting pillars are assymetrically placed....................................... 14 2.1 Example of sampling nine points using three different sampling algorithms: Random, Grid, and Latin Hypercube sampling.................................. 22 xi 3.1 The ADA workflow with an optimizer module. In this workflow, the user only needs to interact with theAD tool and to select the options for the optimizer, enabling the automatic optimization of the design..................................... 31 3.2 An example of the interactive plot feature, where an architect selected the seventh evalu- ation, the red dot on the left plot, and the values of the parameters in that evaluation were used to generate a 3D model using a CAD tool on the right.................. 33 4.1 A plot of the Himmelblau
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages87 Page
-
File Size-